Without simulation, complex systems would fail. Satellites would not reach an accurate orbit, semiconductor circuits would not function, and bridges would not carry the load. Businesses and governments would not invest in these projects without robust simulation software. And without a simulation proving value and functionality, IoT networks of hundreds of thousands or millions of inexpensive devices adding up to large capital investments will not be built.
Researchers from the University of Bologna published an analysis of IoT simulation and a smart cities vehicular transportation system case study (pdf). They recommend a networked simulation of orchestrated simulators that model specific IoT features that fit the diversity of IoT devices and use cases.
According to the researchers, one single simulator is not available to model the many different types of devices. Running on a single CPU is not scalable because of the huge number of events that will be stored and processed. They recommend parallel and distributed simulation (PADS).
The PADS architecture is similar to systems applied to large web applications and artificial intelligence models. Within this distributed simulation, each logical process that implements a model component manages the evolution of the discrete model and communicates with other logical processes for synchronization and data distribution.
Pros and cons of multi-level hybrid simulation
Multi-level hybrid simulation has some advantages and disadvantages. Specific simulators, such as telecommunication modeling tools, could identify congestion caused by the IoT in Wi-Fi and WAN networks. Though, independent builders of component simulators would have to cooperate to some degree so that state information could be shared and synchronized between simulators. Otherwise, low-level events in one simulator would not trigger an event in another.
A multi-level approach to simulation could reduce the use of computational resources by utilizing fine-grained simulation where necessary. For example, fine-grained traffic simulation may be necessary only during peak traffic times for congestion constrained areas. High-level simulation could be effective for off-peak times.
Design, setup and tuning of large-scale simulators depend on either IoT test beds or simulation tools. There are no test beds of sufficient scale, though. If all the test beds in the world were federated and available to researchers, businesses and governments for tuning simulators, the test bed option might work. Otherwise, tools are the only option.
The two most interesting, and perhaps promising approaches are the internet of simulators (IoS) and specification and description language (SDL). An IoS built on interconnected simulators is envisioned as a cloud service available on demand.
The SDL approach is more pragmatic and has been implemented in products such as Telelogic and ObjectGeode applied in many industries. SDLs define systems as a set of interconnected abstract machines, which are extensions of finite state machines. More interesting is the possible nexus of SDLs and system design language.
System design languages, such as VHDL, have proved effective in automating the design and simulation of very large-scale integrated circuits with billions of transistors. The EDA industry standardized on VHDL. It is analogous to a programming language. The system is defined in this high-level language and can be simulated, tested and compiled into an integrated circuit. It promotes reuse and allows models to be exchanged within the ecosystem so that systems of many integrated circuits can be simulated without building a prototype. A similar approach could be taken by the IoT industry, though it may be too early in the evolution of IoT to implement standards for a system design language.
The vehicular transportation system case study is a simulation using a street map to identify the critical points and to tune or modify the traffic circulation in an area. It also considers polluting emissions. Street information represented as a network, including dimensional data and street type (one-way, two-way, highway, etc.) were exported from Open Street Maps. This data was then simulated using network analysis tools NetworkX and Gephi to identify critical points and triggers.
SUMO, a well-known tool for traffic simulation, was used to model the traffic by varying a number of vehicles, assessing traffic circulation when introducing barriers, removing roads or adding novel ones. ADVISOR, a MATLAB-based tool, simulated tailpipe emissions, performance and fuel economy of gasoline/diesel, electric and hybrid vehicles with the intent of reducing tailpipe emissions while improving traffic conditions.
Omnet++ was used to simulate vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications to study intermittent connections, study seamless communications strategies and test multi-homing mechanisms.
Lastly, the performance of the cloud was tested because most IoT networks are implemented as a cloud service. A simulation of communications, devices and sensors would be incomplete if the performance and network latency of the cloud were excluded.
Researchers were able to prove that scalability, parallelization and triggering high-level vs. fine grained simulation worked. They proved that a hybrid simulation that combined many purpose-built simulators also worked. The paper as a whole raises one material question: How many organizations have the expertise to construct a simulator from many complex tools? Organizations considering expensive capital outlays for large IoT networks will need a cloud-based integrated solution. An IoT specification description language would be a start that could produce the models for a system design language simulation system.